A Small Fishing Vessel Recognition Method using Transfer Learning based on Laser Sensors

Abstract

The management of small vessels has always been key to maritime administration. This paper presents a novel method for recognizing small fishing vessels based on laser sensors. Using four types of small fishing vessels as targets, a recognition method for small fishing vessels based on Markov transition field (MTF) time-series images and VGG-16 transfer learning is proposed. In contrast to conventional methods, this study uses polynomial fitting to obtain the contours of a fishing vessel and transforms one-dimensional vessel contours into two-dimensional time-series images using the MTF coding method. The VGG-16 model is used for the recognition process, and migration learning is applied to improve the results. The UCR time-series public dataset is used as a transfer learning dataset for the MTF time-series image encoding. The experiment demonstrates that the proposed method exhibits higher accuracy and performance than 1D-CNN and other general neural network models, and the highest accuracy rate is 98.92%.